← CHI 2026 map

CHI '26 · Best paper · full-paper review · confidence high

PleaSQLarify: Visual Pragmatic Repair for Natural Language Database Querying

Robin Shing Moon Chan , Rita Sevastjanova , Mennatallah El-Assady

DOI PDF Program page Page JSON

PleaSQLarify is compelling because it reframes ambiguity from a parsing failure into an interaction opportunity. The paper tightly connects pragmatic theory, an information-gain-driven repair algorithm, and a visual interface that helps users inspect and steer alternative SQL interpretations instead of accepting a single opaque guess.

Axes Lens

Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.

Review Summary

PleaSQLarify stands out because it does not merely add another clarification widget to text-to-SQL; it offers a coherent interaction model grounded in pragmatic theory and then carries that model through algorithm design, interface design, and evaluation. The central move is to treat underspecification as normal rather than exceptional. Instead of assuming the system should silently choose one best interpretation, the paper argues that ambiguity should be surfaced and collaboratively repaired. That framing matters because it changes what counts as a good interface: not just one that predicts well, but one that helps users understand the space of plausible interpretations and efficiently narrow it. The algorithmic contribution is also more substantive than a generic ranking scheme. Section 5 lays out a pipeline that generates candidate actions, clusters functionally similar outputs, extracts atomic and grouped decision variables, and ranks clarifications by expected information gain. The grouping step is especially important because it tries to bridge internal model distinctions and user-meaningful choices. The interface contribution complements this by exposing the action space, supporting exploration, and keeping updates traceable across turns. That combination gives the paper a strong internal coherence: the theory motivates the algorithm, and the algorithm motivates the interface. The evidence is well matched to the claims. The quantitative evaluation on AMBROSIA supports the claim that clustering-based clarification reduces uncertainty faster than baselines, while the user study shows that participants could recognize alternative interpretations and adopt different navigation strategies. Just as importantly, the paper is explicit about its limits. It assumes a well-formed intent, fixes the candidate pool to around 50 generations, requires some SQL literacy, and presents the interface as a research tool rather than a deployable end-user system. Those caveats narrow the generality of the current implementation, but they also make the contribution more credible. Overall, this is a strong CHI paper because it turns a familiar technical problem into a well-articulated interaction principle: ambiguity should be made visible and repaired with the user, not hidden behind a single prediction.

What Changed

Canon before

Dominant natural language database querying approaches collapse input ambiguity into a single predicted query without supporting effective user clarification or surfacing alternative interpretations, assuming alignment of system and user priors over possible actions.

Departure from common sense

Contrary to typical systems that treat ambiguity as a problem to resolve silently, this paper treats underspecification as an inherent communication feature and proposes incremental, pragmatic repair via interactive clarification that exposes the system's action space and aligns system-user priors.

Actual novelty

The paper introduces a principled framework applying pragmatic repair theory to natural language interfaces, specifically instantiated in PleaSQLarify for text-to-SQL querying. It presents an algorithm that clusters candidate queries, extracts interpretable grouped decision variables prioritized via expected information gain, and a visual interface that surfaces the action space and enables traceable, minimal clarifications that efficiently reduce ambiguity.

Evidence

The paper supports its claims with a coherent mixed-methods package. The introduction and abstract clearly position the departure from one-shot ambiguity resolution toward interactive pragmatic repair. Section 5 specifies the algorithmic novelty: generating candidate actions, clustering functionally similar outputs, extracting grouped decision variables, and ranking them by expected information gain. Section 7 evaluates the method on AMBROSIA with 50 generated candidates per sample and reports faster uncertainty reduction and higher functional coherence than baselines. The limitations section explicitly bounds the contribution by noting assumptions about well-formed intent, a fixed candidate pool, SQL literacy requirements, research-tool status, and scalability constraints.

Limits

Method limits

The algorithm depends on a finite candidate pool and task-specific feature extraction from SQL candidates, so its behavior is tied to the quality of generated programs and the availability of interpretable atomic features. The paper also states that the experimental setup assumes a user with a well-formed intent that can be mapped to a valid SQL query, excluding cases where intent is uncertain or changes during interaction.

Deployment limits

The interface is presented primarily as a research tool rather than an end-user product. It requires basic SQL knowledge to interpret clauses and outputs, and the authors note that larger databases would likely require more clarification steps while making output-based analysis more costly and harder to visualize.

Boundary conditions

The approach is best suited to settings where candidate actions can be sampled in a manageable finite set, users can reason about SQL-level or output-level distinctions, and the task involves clarifying an existing intent rather than discovering or revising one. Performance and usability may degrade as database scale, query complexity, or user unfamiliarity with SQL increase.

Position in field

This work extends pragmatic inference into HCI for text-to-SQL by combining a theory-driven clarification model with an interactive visualization interface. It positions itself against black-box disambiguation pipelines by making alternative interpretations visible, grouping functionally similar candidates, and treating ambiguity resolution as collaborative repair rather than one-shot prediction.

Abstract

Natural language database interfaces broaden data access, yet they remain brittle under input ambiguity. Standard approaches often collapse uncertainty into a single query, offering little support for mismatches between user intent and system interpretation. We reframe this challenge through pragmatic inference: while users economize expressions, systems operate on priors over the action space that may not align with the users'. In this view, pragmatic repair---incremental clarification through minimal interaction---is a natural strategy for resolving underspecification. We present PleaSQLarify, which operationalizes pragmatic repair by structuring interaction around interpretable decision variables that enable efficient clarification. A visual interface complements this by surfacing the action space for exploration, requesting user disambiguation, and making belief updates traceable across turns. In a study with twelve participants, PleaSQLarify helped users recognize alternative interpretations and efficiently resolve ambiguity. Our findings highlight pragmatic repair as a design principle that fosters effective user control in natural language interfaces.